Rule registered (51571a1) BEFORE this metric was computed; see
tasks/chain13_criteria.md ‘L4 v2 rule’ for provenance (the
v1 rule’s window-exit caveat motivated this; v1 stands as recorded). δ:
ze near events vs seeded matched controls; +1 = pointing error higher
near every event than every control sample.
| run | thaw events | thaw δ (adjudicates) | hold crossings | hold δ (info) | A/B sample sizes (thaw) |
|---|---|---|---|---|---|
| r0_6dof_path | 0 | — | 11891 | -0.547 | 0/0 |
| r0_6dof_minus2pct | 0 | — | 349 | -0.119 | 0/0 |
| r0_6dof_plus2pct | 0 | — | 4338 | -0.588 | 0/0 |
| r1_7dof_path | 47 | +0.257 | 30 | +0.612 | 508/5080 |
| r1_7dof_minus2pct | 37 | +0.613 | 6 | +0.768 | 254/2540 |
| r1_7dof_plus2pct | 8 | +0.933 | 16 | +0.917 | 88/880 |
| r2_7dof_reactive | 62 | +0.293 | 38 | +0.370 | 678/6780 |
| r3_7dof_anchor | 59 | +0.480 | 26 | +0.647 | 634/6340 |
Verdict: L4 NO-GO under v2 — thaw δ ≥ +0.5 on 2/5 7-DOF runs (bar ≥ 3/5).
Control-pool honesty note: r0’s 11,891 hold crossings leave a small far-from-any- transition pool — its hold δ is informational only, per the registration.
regen: /Users/antoniahoffman/miniforge3/envs/new-pin-env/bin/python validation/chain13_l4v2.py